I
A
E
S
I
n
t
e
r
n
at
io
n
al
Jou
r
n
al
of
A
r
t
if
ic
ia
l
I
n
t
e
ll
ig
e
n
c
e
(
I
J
-
AI
)
V
ol
. 14, No. 5, O
c
to
be
r
2025
, pp.
3958
~
3969
I
S
S
N
:
2252
-
8938
,
D
O
I
:
10.11591/
ij
a
i.
v
14
.i
5
.pp
3958
-
3969
3958
Jou
r
n
al
h
om
e
page
:
ht
tp
:
//
ij
ai
.
ia
e
s
c
or
e
.c
om
T
r
an
sl
at
i
on
-
b
ase
d
i
m
age
st
e
gan
ogr
ap
h
y sys
t
e
m
u
t
i
l
i
z
i
n
g
au
t
oe
n
c
od
e
r
an
d
C
yc
l
e
G
A
N
T
h
ak
w
an
A
k
r
am
Jaw
ad
1
, Jam
s
h
id
B
agh
e
r
z
ad
e
h
M
oh
as
e
f
i
2
,
M
oh
am
m
e
d
S
al
ah
R
e
d
a A
b
d
e
lg
h
an
y
3
1
D
e
pa
r
t
m
e
nt
of
S
ys
t
e
m
a
nd C
ont
r
ol
E
ngi
ne
e
r
i
ng, C
ol
l
e
ge
of
E
l
e
c
t
r
oni
c
E
ng
i
ne
e
r
i
ng, N
i
ne
va
h U
ni
ve
r
s
i
t
y,
M
os
ul
, I
r
a
q
2
D
e
pa
r
t
m
e
nt
of
C
om
put
e
r
E
ngi
ne
e
r
i
ng, F
a
c
ul
t
y of
E
l
e
c
t
r
i
c
a
l
a
nd C
om
put
e
r
E
ng
i
ne
e
r
i
ng, U
r
m
i
a
U
ni
ve
r
s
i
t
y, U
r
m
i
a
, I
r
a
n
3
D
e
pa
r
t
m
e
nt
of
C
om
put
e
r
s
a
nd A
r
t
i
f
i
c
i
a
l
I
nt
e
l
l
i
ge
nc
e
, D
a
m
i
e
t
t
a
U
ni
ve
r
s
i
t
y, D
a
m
i
e
t
t
a
, E
gypt
A
r
t
ic
le
I
n
f
o
A
B
S
T
R
A
C
T
A
r
ti
c
le
h
is
to
r
y
:
R
e
c
e
iv
e
d
A
ug 24, 2024
R
e
vi
s
e
d
J
un 16, 2025
A
c
c
e
pt
e
d
J
ul
10, 2025
Traditional
image
steganography
involves
embedding
secret
infor
mation
into
a
cover
image,
a
process
that
requires
modification
of
the
carri
er
and
potentially
leaves
detectable
marks.
This
paper
proposes
a
novel
met
hod
of
coverless
image
steganogr
aphy
based
on
generative
models.
Initial
ly,
a
CycleGA
N
model
is
construc
ted
and
trained
to
learn
the
featur
es
of
di
ffere
nt
image
domains.
Subsequently,
an
Autoencoder
model
is
trained
usi
ng
two
sets of images to a
chieve a pr
ecise
one
-
to
-
one mapping. Once the mod
els
are
trained,
the
autoencoder
is
used
on
both
the
sender
and
re
ceiver
si
des
to
convert
the
cover
image
(also
known
as
the
stego
image)
into
the
secret
image
and
vice
versa.
The
CycleGAN
model
is
then
utilized
to
enha
nce
the
visual
quality
of
the
images
generated
by
the
autoencoder
.
Experi
mental
results
demonstrate
that
this
method
not
only
effectively
secures
secret
information
transmission
but
also
improves
efficie
ncy
and
increas
es
the
capacity fo
r informat
ion hi
ding com
pared to
simil
ar method
s.
K
e
y
w
o
r
d
s
:
A
ut
oe
nc
ode
r
C
ove
r
le
s
s
s
te
ga
nogr
a
phy
C
yc
le
G
A
N
G
e
ne
r
a
ti
ve
m
ode
ls
S
te
ga
nogr
a
phy
This is an
open
acce
ss artic
le unde
r the
CC BY
-
SA
license.
C
or
r
e
s
pon
di
n
g A
u
th
or
:
J
a
m
s
hi
d B
a
gh
e
r
z
a
de
h
M
oha
s
e
f
i
D
e
pa
r
tm
e
nt
of
C
om
put
e
r
E
ngi
ne
e
r
in
g, F
a
c
ul
ty
of
E
le
c
tr
ic
a
l
a
nd C
om
put
e
r
E
ngi
ne
e
r
in
g, U
r
m
ia
U
ni
ve
r
s
it
y
11
km
S
e
r
ow
R
oa
d, U
r
m
ia
, W
e
s
t
A
z
e
r
ba
ij
a
n, I
r
a
n
E
m
a
il
:
j.
ba
ghe
r
z
a
de
h@
ur
m
ia
.a
c
.i
r
1.
I
N
T
R
O
D
U
C
T
I
O
N
S
te
ga
n
og
r
a
p
hy
is
th
e
s
c
i
e
nc
e
a
nd
a
r
t
o
f
m
a
ki
ng
t
he
in
f
o
r
m
a
t
io
n
in
vi
s
ib
l
e
by
us
in
g
di
f
f
e
r
e
nt
c
onc
e
a
li
ng
m
e
th
ods
.
S
im
il
a
r
t
o
c
r
y
pt
og
r
a
p
hy,
s
te
ga
n
og
r
a
p
h
y
t
r
ie
s
to
e
ns
u
r
e
th
e
s
e
c
r
e
c
y
a
nd
s
a
f
e
ty
o
f
in
f
o
r
m
a
ti
on
,
b
ut
u
nl
ik
e
c
r
yp
to
gr
a
ph
y,
s
te
g
a
no
g
r
a
ph
y
t
r
ie
s
to
hi
de
t
he
e
xi
s
te
n
c
e
of
t
hi
s
s
e
c
r
e
t
in
f
or
m
a
t
io
n
[
1
]
.
T
he
m
os
t
w
e
ll
-
k
no
w
n
e
m
be
dd
in
g
a
lg
o
r
it
h
m
is
th
e
le
a
s
t
s
ig
n
i
f
ic
a
nt
bi
t
(
L
S
B
)
a
l
go
r
i
th
m
.
T
he
id
e
a
o
f
L
S
B
is
to
e
nc
od
e
th
e
s
e
c
r
e
t
m
e
s
s
a
ge
in
to
t
he
L
S
B
o
f
a
c
o
lo
r
c
ha
nne
l
in
th
e
c
o
ve
r
im
a
ge
.
S
i
nc
e
t
he
L
S
B
a
lg
o
r
it
hm
m
a
ni
pul
a
te
s
t
he
pi
x
e
l
i
n
de
p
e
nd
e
nt
ly
o
f
e
a
c
h
o
th
e
r
,
i
t
is
v
ul
ne
r
a
b
le
to
s
te
g
a
na
ly
s
is
t
e
c
h
ni
que
s
a
nd
pr
one
to
de
te
c
ti
on
[
2]
.
T
w
o
c
r
uc
ia
l
r
e
qui
r
e
m
e
nt
s
of
e
v
e
r
y
tr
a
di
ti
ona
l
s
te
ga
nogr
a
phy
s
ys
te
m
a
r
e
th
e
d
a
ta
a
nd
a
c
a
r
r
ie
r
.
A
c
a
r
r
ie
r
is
r
e
f
e
r
r
e
d
to
th
e
pa
pe
r
,
im
a
g
e
,
vi
de
o,
or
a
ny
m
ul
ti
m
e
di
a
th
a
t
c
a
r
r
ie
s
th
e
s
e
c
r
e
t
da
t
a
.
T
h
e
da
ta
is
e
m
be
dde
d
in
to
th
e
c
a
r
r
ie
r
,
a
nd
th
e
n
tr
a
ns
m
it
te
d
to
r
e
c
ip
ie
nt
.
H
ow
e
ve
r
,
th
is
m
e
th
od
r
a
is
e
s
s
om
e
is
s
u
e
s
,
be
c
a
us
e
th
e
s
e
c
r
e
t
da
ta
is
e
m
be
dde
d
in
to
th
e
c
ove
r
,
f
or
in
s
ta
nc
e
by
m
a
ni
pul
a
ti
ng
th
e
pi
xe
ls
of
th
e
c
a
r
r
ie
r
im
a
ge
,
a
tt
a
c
ke
r
c
a
n
de
te
c
t
th
e
e
xi
s
te
n
c
e
of
a
s
e
c
r
e
t
m
e
s
s
a
ge
in
th
e
im
a
ge
or
a
ny
m
e
di
um
it
is
be
in
g
tr
a
ns
m
it
te
d w
it
h [
1]
,
[
3]
–
[
5]
.
S
te
ga
na
ly
s
is
,
a
n
e
m
e
r
gi
ng
f
ie
ld
of
s
tu
dy
pa
r
a
ll
e
l
to
s
te
ga
nogr
a
phy,
is
r
e
f
e
r
r
e
d
to
th
e
s
c
ie
nc
e
of
de
te
c
ti
ng
th
e
e
xi
s
te
nc
e
of
hi
dde
n
da
ta
in
th
e
c
ove
r
f
il
e
.
T
he
a
ppr
oa
c
he
s
ut
il
iz
e
d
f
or
s
te
ga
na
ly
s
is
s
om
e
ti
m
e
s
de
pe
nd
on
th
e
s
te
ga
nogr
a
phy
a
lg
or
it
hm
(
s
)
us
e
d
to
c
onc
e
a
l
th
e
da
ta
[
6]
.
I
n
or
de
r
to
a
ddr
e
s
s
th
e
c
onc
e
r
ns
w
it
h
th
e
c
onve
nt
io
na
l
s
t
e
ga
nogr
a
phy,
e
xpe
r
ts
pr
opos
e
d
c
ov
e
r
le
s
s
in
f
or
m
a
ti
on
hi
di
ng
in
2014.
C
ove
r
le
s
s
Evaluation Warning : The document was created with Spire.PDF for Python.
I
nt
J
A
r
ti
f
I
nt
e
ll
I
S
S
N
:
2252
-
8938
T
r
ans
la
ti
on
-
bas
e
d i
m
age
s
te
ganog
r
aphy
s
y
s
te
m
ut
il
iz
in
g autoe
nc
ode
r
and …
(
T
hak
w
an A
k
r
am
J
aw
ad)
3959
in
f
or
m
a
ti
on
hi
di
ng
is
r
e
f
e
r
r
e
d
to
th
e
na
tu
r
a
l
c
a
r
r
ie
r
,
w
hi
c
h
is
c
om
pe
ll
e
d
by
th
e
s
e
c
r
e
t
da
ta
.
B
y
s
ha
r
in
g
th
e
m
a
ppi
ng
be
twe
e
n
c
e
r
ta
in
f
e
a
tu
r
e
s
of
th
e
c
a
r
r
ie
r
a
nd
th
e
s
e
c
r
e
t
da
ta
,
s
e
nde
r
a
nd
r
e
c
e
iv
e
r
c
a
n
c
om
m
uni
c
a
te
s
e
c
r
e
t
in
f
or
m
a
ti
on w
it
hout
c
ha
ngi
ng or
m
a
ni
pul
a
ti
ng t
he
m
e
di
um
a
ls
o known a
s
t
he
c
ove
r
[
5]
.
G
e
ne
r
a
ti
ve
a
dve
r
s
a
r
ia
l
ne
twor
ks
(
G
A
N
)
a
r
e
a
va
r
ia
ti
on
of
de
e
p
c
onvolut
io
na
l
ne
ur
a
l
ne
twor
k
(
C
N
N
s
)
put
f
or
th
by
A
lm
a
ha
ir
i
e
t
al
.
[
7
]
.
A
G
A
N
is
m
a
de
of
t
w
o
de
e
p
ne
twor
ks
ge
ne
r
a
to
r
a
nd
di
s
c
r
im
in
a
to
r
c
om
pe
te
a
ga
in
s
t
e
a
c
h
ot
he
r
in
a
z
e
r
o
-
s
um
ga
m
e
to
pr
oduc
e
a
va
li
d
out
put
in
G
A
N
a
r
c
hi
te
c
tu
r
e
. T
he
ge
ne
r
a
to
r
ne
twor
k
is
tr
a
in
e
d
in
a
w
a
y
to
p
r
oduc
e
s
im
il
a
r
im
a
ge
s
t
o
it
s
in
put
a
nd
f
ool
th
e
di
s
c
r
im
in
a
to
r
.
T
he
di
s
c
r
im
in
a
to
r
is
tr
a
in
e
d
to
f
in
d
th
e
f
a
ke
im
a
g
e
s
e
f
f
e
c
ti
ve
ly
[
8]
.
G
e
ne
r
a
ll
y,
a
G
A
N
m
ode
l
is
m
a
de
of
two
c
hi
e
f
c
om
pone
nt
s
.
W
hi
c
h
a
r
e
di
s
c
r
im
in
a
to
r
a
nd
ge
n
e
r
a
to
r
.
A
ne
w
ne
twor
k
na
m
e
d
th
e
s
te
ga
na
ly
z
e
r
,
w
it
h
th
e
pur
pos
e
of
c
he
c
ki
ng i
f
t
he
i
nput
ha
s
a
ny c
onc
e
a
le
d da
t
a
i
n i
t
or
not
, i
s
us
e
d i
n s
om
e
m
e
th
ods
w
it
hi
n t
he
i
m
a
ge
s
te
ga
nogr
a
phy c
ont
e
xt
[
8]
.
C
yc
le
G
A
N
,
a
bb
r
e
v
ia
te
d
f
r
om
c
y
c
le
-
c
o
ns
is
te
n
t
ge
n
e
r
a
t
iv
e
a
dve
r
s
a
r
ia
l
ne
t
w
o
r
ks
,
is
a
ge
ne
r
a
ti
ve
m
ode
l
e
m
pl
oye
d
in
c
o
m
p
ut
e
r
v
is
i
on
a
n
d
im
a
ge
s
yn
th
e
s
is
.
I
t
w
a
s
de
vi
s
e
d
to
f
a
c
il
it
a
te
uns
upe
r
v
is
e
d
le
a
r
n
in
g
f
o
r
i
m
a
ge
-
to
-
i
m
a
ge
tr
a
ns
la
ti
on
,
e
l
im
in
a
ti
ng
t
he
ne
e
d
f
o
r
pa
i
r
e
d
tr
a
in
in
g
da
t
a
[
9
]
.
I
n
t
he
C
yc
le
G
A
N
f
r
a
m
e
w
o
r
k,
a
ge
ne
r
a
t
or
is
tr
a
in
e
d
t
o
p
r
o
duc
e
i
m
a
ge
s
in
one
d
om
a
in
b
a
s
e
d
on
im
a
ge
s
f
r
om
a
no
th
e
r
d
om
a
in
.
S
i
nc
e
th
e
r
e
is
no
r
e
l
ia
n
c
e
on
pa
i
r
e
d
i
n
f
o
r
m
a
ti
on
,
nu
m
e
r
ou
s
po
te
nt
ia
l
m
a
pp
in
gs
c
ou
ld
b
e
in
f
e
r
r
e
d.
T
o
c
ons
t
r
a
in
th
e
m
ul
ti
tu
d
e
o
f
p
os
s
i
bl
e
m
a
ppi
ngs
,
C
yc
le
G
A
N
is
c
om
m
o
nl
y
t
r
a
in
e
d
w
i
th
a
c
y
c
le
-
c
ons
is
te
n
c
y
c
ons
t
r
a
in
t.
T
hi
s
c
ons
tr
a
i
nt
e
ns
u
r
e
s
a
r
ob
us
t
c
on
ne
c
t
io
n
a
c
r
os
s
d
om
a
in
s
by
m
a
n
da
ti
n
g
th
a
t
t
he
tr
a
ns
f
o
r
m
a
ti
on
o
f
a
n
i
m
a
ge
f
r
o
m
th
e
s
ou
r
c
e
do
m
a
i
n
to
th
e
t
a
r
ge
t
d
om
a
in
a
nd
b
a
c
k
to
t
he
s
o
ur
c
e
s
ho
ul
d
yi
e
ld
th
e
o
r
i
gi
na
l
i
m
a
g
e
[
7
]
.
I
n
our
r
e
s
e
a
r
c
h,
w
e
w
il
l
dr
a
w
in
s
pi
r
a
ti
on
f
r
om
nove
l
c
onc
e
pt
s
pr
e
s
e
nt
e
d
in
th
e
s
tu
di
e
s
c
onduc
te
d
pr
io
r
to
th
is
r
e
s
e
a
r
c
h,
s
pe
c
if
ic
a
ll
y
a
dopt
in
g
f
e
a
tu
r
e
m
a
ppi
ng
t
e
c
hni
que
.
H
ow
e
ve
r
,
a
s
ig
ni
f
ic
a
nt
de
pa
r
tu
r
e
in
our
pr
opos
e
d
s
te
ga
nogr
a
phy
s
y
s
te
m
is
it
s
f
ul
ly
-
c
ove
r
le
s
s
na
tu
r
e
.
W
he
n
c
om
pa
r
e
d
to
tr
a
di
ti
ona
l
s
te
ga
nogr
a
phi
c
m
e
th
ods
, our
c
ove
r
le
s
s
s
te
ga
nogr
a
phy s
ol
ut
io
n
e
xhi
bi
ts
s
e
ve
r
a
l
a
dv
a
nt
a
ge
s
:
‒
I
m
pe
r
c
e
pt
ib
il
it
y:
t
r
a
di
ti
ona
l
m
e
th
ods
of
te
n
le
a
ve
de
te
c
ta
bl
e
tr
a
c
e
s
,
m
a
ki
ng
th
e
m
s
us
c
e
pt
ib
le
to
s
te
ga
na
ly
s
i
s
.
O
ur
m
e
th
od,
how
e
ve
r
,
pr
oduc
e
s
im
a
g
e
s
th
a
t
a
r
e
vi
r
tu
a
ll
y
in
di
s
ti
ngui
s
ha
bl
e
f
r
om
ge
nui
ne
im
a
ge
s
, s
ig
ni
f
ic
a
nt
ly
e
nha
nc
in
g i
m
pe
r
c
e
pt
ib
il
it
y.
‒
H
ig
he
r
da
ta
c
a
p
a
c
it
y:
by
not
e
m
be
ddi
ng
da
ta
di
r
e
c
tl
y
in
to
th
e
i
m
a
ge
pi
xe
ls
,
our
m
e
th
od
c
ir
c
um
ve
nt
s
th
e
c
a
pa
c
it
y l
im
it
a
ti
ons
of
t
r
a
di
ti
ona
l
te
c
hni
que
s
, a
ll
ow
in
g f
or
m
or
e
s
ubs
ta
nt
ia
l
in
f
or
m
a
ti
on t
o be
c
onc
e
a
le
d.
‒
E
f
f
ic
ie
nc
y:
th
e
us
e
of
a
ut
oe
nc
ode
r
s
a
nd
C
yc
le
G
A
N
a
ll
ow
s
f
or
e
f
f
ic
ie
nt
da
ta
c
onc
e
a
lm
e
nt
a
nd
r
e
tr
ie
va
l
pr
oc
e
s
s
e
s
.
T
hi
s
e
f
f
ic
ie
nc
y
is
c
r
it
ic
a
l
f
or
r
e
a
l
-
ti
m
e
a
ppl
ic
a
ti
o
ns
a
nd
s
c
e
na
r
io
s
w
he
r
e
c
om
put
a
ti
ona
l
r
e
s
our
c
e
s
m
a
y be
l
im
it
e
d.
T
hi
s
pa
pe
r
is
s
tr
uc
tu
r
e
d
a
s
f
ol
lo
w
s
:
t
he
s
e
c
ond
s
e
c
ti
on
pr
ovi
de
s
a
c
om
pr
e
he
ns
iv
e
r
e
vi
e
w
of
pr
io
r
r
e
s
e
a
r
c
h
in
th
e
f
ie
ld
,
e
s
ta
bl
is
hi
ng
th
e
c
ont
e
xt
f
or
th
is
s
tu
dy.
T
he
th
ir
d
s
e
c
ti
on
de
ta
il
s
our
pr
opos
e
d
m
e
th
od,
out
li
ni
ng
it
s
de
s
ig
n
a
nd
im
pl
e
m
e
nt
a
ti
on.
I
n
th
e
f
our
th
s
e
c
ti
on,
w
e
e
va
lu
a
te
th
e
e
f
f
e
c
ti
ve
ne
s
s
of
our
a
ppr
oa
c
h
th
r
ough
two
c
a
r
e
f
ul
ly
de
s
ig
ne
d
e
xpe
r
im
e
nt
a
l
s
c
e
n
a
r
io
s
.
F
in
a
ll
y,
th
e
f
if
th
s
e
c
ti
on
c
onc
lu
de
s
th
e
pa
pe
r
,
s
um
m
a
r
iz
in
g ke
y f
in
di
ngs
a
nd i
m
pl
ic
a
ti
ons
of
t
hi
s
r
e
s
e
a
r
c
h.
2.
R
E
L
A
T
E
D
WORK
G
A
N
a
r
e
va
r
ia
ti
on
s
of
d
e
e
p C
N
N
s
[
10]
,
[
11]
put
f
or
th
by
G
oodf
e
ll
ow
e
t
al
.
[
12]
.
A
G
A
N
is
m
a
de
of
two
de
e
p
ne
twor
ks
ge
ne
r
a
to
r
a
nd
di
s
c
r
im
in
a
to
r
c
om
pe
te
a
ga
in
s
t
e
a
c
h
ot
he
r
in
a
z
e
r
o
-
s
um
ga
m
e
to
pr
oduc
e
a
va
li
d
im
a
ge
in
G
A
N
a
r
c
hi
te
c
tu
r
e
.
T
h
e
ge
ne
r
a
to
r
ne
twor
k
is
tr
a
in
e
d
in
a
w
a
y
to
pr
oduc
e
s
im
il
a
r
im
a
ge
s
to
it
s
in
put
a
nd
f
ool
th
e
di
s
c
r
im
in
a
to
r
,
a
nd
th
e
di
s
c
r
im
in
a
to
r
is
tr
a
in
e
d
to
f
in
d
th
e
f
a
ke
im
a
ge
s
e
f
f
e
c
ti
ve
ly
[
13]
.
G
e
ne
r
a
ll
y,
a
G
A
N
m
ode
l
is
m
a
de
of
two
c
hi
e
f
c
om
pone
nt
s
.
W
hi
c
h
a
r
e
di
s
c
r
im
in
a
to
r
a
nd
ge
ne
r
a
to
r
.
A
ne
w
ne
twor
k
na
m
e
d
th
e
s
te
ga
na
ly
z
e
r
,
w
it
h
th
e
pur
pos
e
of
c
he
c
ki
ng
if
th
e
in
put
ha
s
a
ny
c
onc
e
a
le
d
da
ta
or
not
,
is
us
e
d
in
s
om
e
m
e
th
ods
w
it
hi
n
th
e
im
a
ge
s
te
ga
nogr
a
phy
c
ont
e
x
t
[
13]
.
T
he
r
e
a
r
e
di
f
f
e
r
e
nt
va
r
ia
ti
ons
of
G
A
N
be
in
g
us
e
d
in
s
te
ga
nogr
a
phy
ta
s
ks
.
T
o
na
m
e
s
om
e
,
C
yc
le
G
A
N
,
c
ondi
ti
ona
l
g
e
ne
r
a
ti
ve
a
dve
r
s
a
r
ia
l
ne
twor
ks
(
C
G
A
N
)
[
14
]
,
de
e
p
c
onvolut
io
na
l
g
e
ne
r
a
ti
ve
a
dve
r
s
a
r
ia
l
ne
t
w
or
ks
(
D
C
G
A
N
)
,
a
nd
W
a
s
s
e
r
s
te
in
g
e
n
e
r
a
ti
ve
a
dve
r
s
a
r
ia
l
ne
twor
ks
(
W
G
A
N
)
[
15]
.
H
u
e
t
al
.
[
16]
p
r
opos
e
d
th
e
D
C
G
A
N
a
r
c
hi
te
c
tu
r
e
u
s
e
d
f
or
s
te
ga
nogr
a
phy
w
it
hout
e
m
be
ddi
ng
(
S
W
E
)
.
T
he
ir
pr
opos
e
d
m
e
th
od
e
li
m
in
a
te
s
th
e
e
m
be
ddi
ng
pr
o
c
e
s
s
by
g
e
ne
r
a
ti
ng
th
e
c
a
r
r
ie
r
im
a
ge
ba
s
e
d
on
th
e
noi
s
e
ve
c
to
r
w
hi
c
h s
e
c
r
e
t
in
f
or
m
a
ti
on w
a
s
m
a
ppe
d t
o.
I
n t
hi
s
m
e
th
od a
not
he
r
ne
twor
k c
a
ll
e
d t
he
e
xt
r
a
c
to
r
ne
twor
k i
s
r
e
qui
r
e
d t
o r
e
-
a
c
qui
r
e
t
he
s
e
c
r
e
t
in
f
or
m
a
ti
on f
r
om
t
h
e
c
a
r
r
ie
r
i
m
a
ge
.
W
hi
le
s
om
e
s
te
ga
nogr
a
phi
c
a
ppr
oa
c
he
s
ut
il
iz
e
a
s
in
gl
e
ge
n
e
r
a
ti
ve
m
ode
l,
L
i
e
t
al
.
[
17]
pr
opos
e
d
a
two
-
s
ta
ge
m
e
th
od
w
it
h
s
e
pa
r
a
te
m
ode
l
s
(
F
a
nd
G
)
f
or
c
ove
r
im
a
ge
ge
ne
r
a
ti
on
a
nd
s
e
c
r
e
t
im
a
ge
r
e
c
ons
tr
uc
ti
on.
H
ow
e
ve
r
,
th
e
ir
w
or
k
a
c
knowle
dge
s
c
ha
ll
e
nge
s
in
r
e
c
ons
tr
uc
ti
ng
th
e
s
e
c
r
e
t
im
a
ge
due
to
a
la
c
k
of
c
ont
e
nt
in
f
or
m
a
ti
on
pr
e
s
e
r
ve
d
dur
in
g
th
e
in
it
ia
l
c
ove
r
i
m
a
ge
ge
ne
r
a
ti
on
s
te
p
by
m
ode
l
F
.
T
o
a
ddr
e
s
s
th
is
is
s
ue
,
th
e
a
ut
hor
s
in
tr
oduc
e
a
nov
e
l
"
c
ont
e
nt
-
c
ons
is
te
n
c
y
"
e
xt
r
a
c
ti
on
m
odul
e
w
it
hi
n
th
e
c
ove
r
im
a
g
e
ge
ne
r
a
ti
on pr
oc
e
s
s
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2252
-
8938
I
nt
J
A
r
ti
f
I
nt
e
ll
,
V
ol
. 14, No. 5, O
c
to
be
r
2025
:
3958
-
3969
3960
L
i
e
t
al
.
[
18]
in
c
or
por
a
te
d
a
D
C
G
A
N
a
r
c
hi
te
c
tu
r
e
f
or
th
e
di
s
c
r
im
in
a
to
r
.
I
n
th
e
ir
s
tu
dy,
th
e
y
ha
ve
bui
lt
a
m
e
th
od t
ha
t
tr
a
ns
f
e
r
s
s
e
c
r
e
t
im
a
ge
s
be
twe
e
n t
w
o doma
in
s
. T
he
ir
w
or
k pr
opos
e
s
a
t
w
o
-
s
ta
ge
ge
n
e
r
a
ti
ve
m
ode
l
a
ppr
oa
c
h
f
or
s
e
c
r
e
t
im
a
ge
e
nc
r
ypt
io
n.
I
n
th
e
f
ir
s
t
s
ta
ge
,
th
e
y
e
m
be
d
th
e
s
e
c
r
e
t
im
a
ge
w
it
hi
n
a
publ
ic
im
a
ge
f
r
om
a
di
f
f
e
r
e
nt
doma
in
, r
e
s
ul
ti
ng i
n a
s
ynt
he
ti
c
i
m
a
ge
.
T
hi
s
s
ynt
he
ti
c
i
m
a
ge
i
s
t
he
n u
s
e
d a
s
i
nput
f
or
a
ge
ne
r
a
ti
ve
m
ode
l
(
F
)
t
o ge
ne
r
a
te
a
n e
nc
r
ypt
e
d i
m
a
ge
i
n a
not
he
r
doma
in
.
Z
ha
ng
e
t
al
.
[
19]
p
r
opos
e
d
a
ge
ne
r
a
ti
ve
r
e
ve
r
s
ib
le
da
ta
hi
di
n
g
(
G
R
D
H
)
a
ppr
oa
c
h
ut
il
iz
in
g
i
m
a
ge
tr
a
ns
la
ti
on.
T
he
ir
m
e
th
od
in
vol
ve
s
two
s
ta
ge
s
.
F
ir
s
t,
a
n
im
a
ge
ge
ne
r
a
to
r
c
r
e
a
te
s
a
r
e
a
li
s
ti
c
im
a
ge
th
a
t
s
e
r
ve
s
a
s
in
put
to
a
C
yc
le
G
A
N
m
ode
l.
C
yc
le
G
A
N
pe
r
f
o
r
m
s
im
a
ge
-
to
-
im
a
ge
tr
a
ns
la
ti
on,
r
e
s
ul
ti
ng
in
a
s
te
go
-
im
a
ge
th
a
t
c
onc
e
a
l
s
th
e
s
e
c
r
e
t
m
e
s
s
a
ge
.
N
ot
a
bl
y,
bot
h
th
e
s
e
c
r
e
t
m
e
s
s
a
ge
a
nd
th
e
or
ig
in
a
l
im
a
ge
c
a
n
be
r
e
c
ov
e
r
e
d
in
de
pe
nde
nt
ly
.
A
tr
a
in
e
d
m
e
s
s
a
ge
e
xt
r
a
c
to
r
r
e
tr
ie
ve
s
th
e
hi
d
de
n
m
e
s
s
a
ge
,
w
hi
le
th
e
in
ve
r
s
e
of
th
e
im
a
ge
tr
a
ns
la
ti
on pr
oc
e
s
s
r
e
c
ov
e
r
s
t
he
or
ig
in
a
l
im
a
ge
.
D
ua
n
a
nd
S
ong
e
t
al
.
[
13]
in
tr
oduc
e
d
a
nove
l
c
ove
r
le
s
s
im
a
g
e
in
f
or
m
a
ti
on
hi
di
ng
m
e
th
od
ut
il
iz
in
g
ge
ne
r
a
ti
ve
m
ode
l
da
ta
ba
s
e
.
T
h
e
y
pr
opos
e
tr
a
ns
m
it
ti
ng
a
ne
w
ly
ge
ne
r
a
te
d
im
a
ge
,
in
de
pe
nde
nt
f
r
om
th
e
s
e
c
r
e
t
im
a
ge
,
th
a
t
c
a
n
be
de
c
ode
d
ba
c
k
to
th
e
or
ig
in
a
l
s
e
c
r
e
t
im
a
ge
by
th
e
r
e
c
e
iv
e
r
us
in
g
th
e
ir
ge
ne
r
a
ti
ve
m
ode
l
da
ta
ba
s
e
.
T
hi
s
m
e
th
od of
f
e
r
s
i
m
pr
ove
d s
e
c
ur
it
y by tr
a
ns
m
it
ti
ng
a
n uninf
or
m
a
ti
ve
i
m
a
ge
. T
he
y ut
il
iz
e
W
G
A
N
m
ode
l
to
a
c
hi
e
ve
c
ove
r
le
s
s
im
a
ge
in
f
or
m
a
ti
on
hi
di
ng.
I
ns
te
a
d
of
di
r
e
c
tl
y
t
r
a
ns
m
it
ti
ng
th
e
s
e
c
r
e
t
im
a
ge
,
th
e
y
tr
a
in
th
e
W
G
A
N
on
th
e
s
e
c
r
e
t
im
a
ge
(
r
e
pl
a
c
in
g
th
e
us
ua
l
r
a
ndom
noi
s
e
in
put
)
.
T
hi
s
W
G
A
N
c
a
n
th
e
n
ge
ne
r
a
te
a
n
e
w
,
in
de
p
e
nde
nt
"
m
e
a
ni
ng
-
nor
m
a
l"
im
a
ge
th
a
t
doe
s
n'
t
c
ont
a
in
th
e
s
e
c
r
e
t
in
f
or
m
a
ti
on.
T
hi
s
pr
oduc
e
d
pi
c
tu
r
e
is
tr
a
ns
m
it
te
d
to
th
e
r
e
c
e
iv
e
r
,
w
hi
c
h
u
s
e
s
W
G
A
N
m
ode
l
to
r
e
ge
r
ne
r
a
te
th
e
or
ig
in
a
l
s
e
c
r
e
t
im
a
ge
.
H
ow
e
ve
r
,
th
i
s
m
e
th
od
c
onf
r
ont
s
s
c
a
la
bi
li
ty
c
ha
ll
e
ng
e
s
.
T
h
e
c
ur
r
e
nt
a
ppr
oa
c
h
im
pos
e
s
s
to
r
in
g
a
nd
m
a
na
gi
ng
two
s
e
pa
r
a
te
W
G
A
N
m
ode
ls
f
or
e
a
c
h
s
e
nde
r
-
r
e
c
e
iv
e
r
s
e
t.
T
hi
s
c
a
n
be
c
om
e
bul
ky
f
or
la
r
ge
-
s
c
a
le
de
pl
oym
e
nt
s
or
s
c
e
na
r
io
s
w
it
h f
r
e
que
nt
c
om
m
uni
c
a
ti
on pa
r
tn
e
r
c
ha
nge
s
. A
ddi
ti
ona
ll
y, t
he
c
om
put
a
ti
ona
l
c
os
t
li
nke
d w
it
h t
r
a
in
in
g a
nd ma
in
ta
in
in
g t
he
s
e
m
ode
ls
c
oul
d be
a
l
im
it
in
g i
s
s
ue
.
S
ta
r
G
A
N
ha
s
be
e
n
in
tr
oduc
e
d
by
C
hoi
e
t
al
.
[
20]
,
w
hi
c
h
is
a
G
A
N
th
a
t
ha
s
th
e
a
bi
li
ty
to
m
u
lt
i
-
dom
a
in
im
a
ge
tr
a
ns
f
or
m
a
ti
on
th
a
t
c
a
n
be
us
e
d
to
a
c
hi
e
ve
f
a
c
ia
l
a
tt
r
ib
ut
e
m
a
ni
pul
a
ti
on
w
it
hi
n
th
e
c
ove
r
im
a
ge
.
I
n
th
is
ne
twor
k,
th
e
tr
a
in
in
g
pr
oc
e
s
s
c
om
pr
is
e
s
bot
h
th
e
ge
ne
r
a
to
r
ne
twor
k
(
G
)
a
nd
a
di
s
c
r
im
in
a
to
r
ne
twor
k
(
D
)
;
th
e
ge
ne
r
a
to
r
ta
ke
s
a
ta
r
ge
t
dom
a
in
ta
g
th
a
t
r
e
pr
e
s
e
nt
s
th
e
de
s
ir
e
d
a
tt
r
ib
ut
e
s
a
nd
th
e
in
put
im
a
ge
.
A
f
te
r
th
a
t,
a
ne
w
im
a
ge
w
il
l
be
ge
ne
r
a
te
d
to
in
c
or
por
a
te
th
e
s
e
s
pe
c
if
ie
d
a
tt
r
ib
ut
e
s
.
F
or
th
e
di
s
c
r
im
in
a
to
r
ne
twor
k,
it
ha
s
two
f
unc
ti
ons
w
h
e
r
e
th
e
y
di
f
f
e
r
e
n
te
be
twe
e
n
r
e
a
l
a
nd
f
a
ke
im
a
ge
s
in
a
ddi
ti
on
to
r
e
c
ogni
z
in
g
th
e
dom
a
in
(
a
tt
r
ib
ut
e
s
)
a
s
s
oc
ia
te
d
w
it
h
th
e
in
put
i
m
a
ge
.
A
c
om
bi
na
ti
on
of
lo
s
s
f
unc
ti
ons
w
il
l
b
e
hol
d by S
ta
r
G
A
N
t
o a
c
hi
e
ve
t
he
s
e
goa
ls
, t
hi
s
i
nc
lu
de
s
a
dve
r
s
a
r
ia
l,
c
la
s
s
if
ic
a
ti
on, a
nd r
e
c
ons
tr
uc
ti
on l
os
s
e
s
.
B
y
e
m
pl
oyi
ng
a
n
or
ig
in
a
l
f
o
r
m
of
G
A
N
a
r
c
hi
te
c
tu
r
e
,
S
hi
e
t
al
.
[
21]
d
e
vi
a
te
s
f
r
om
e
xi
s
ti
ng
te
c
hni
que
s
in
th
e
ir
pr
opos
e
d
m
e
th
od
th
a
t
c
ons
is
t
s
of
on
e
ge
ne
r
a
ti
ve
ne
twor
k
a
nd
two
di
s
c
r
im
in
a
ti
ve
ne
twor
ks
.
T
he
r
ol
e
of
th
e
ge
ne
r
a
ti
ve
ne
twor
k
is
to
pr
io
r
it
iz
e
s
th
e
vi
s
ua
l
qua
li
ty
of
th
e
s
te
go
im
a
g
e
s
,
w
hi
l
e
th
e
r
ol
e
of
th
e
is
c
r
im
in
a
ti
ve
ne
twor
ks
is
to
a
s
s
e
s
s
th
e
ir
s
ui
ta
bi
li
t
y
f
or
in
f
or
m
a
ti
on
hi
di
ng.
T
he
a
ut
hor
s
de
c
la
r
e
c
ons
id
e
r
a
bl
e
im
pr
ove
m
e
nt
s
in
th
e
s
pe
e
d
of
c
onve
r
ge
nc
e
,
s
t
a
bi
li
ty
of
th
e
tr
a
in
in
g,
a
nd
th
e
qu
a
li
ty
of
th
e
im
a
ge
.
A
ls
o,
th
e
y
us
e
d
a
n
a
dva
nc
e
d
s
te
ga
na
ly
s
is
ne
twor
k
w
it
hi
n
th
e
di
s
c
r
im
in
a
ti
ve
s
tr
uc
tu
r
e
;
th
is
e
na
bl
e
s
be
tt
e
r
e
va
lu
a
ti
on
of
th
e
ge
ne
r
a
te
d
im
a
ge
s
'
p
e
r
f
or
m
a
nc
e
.
I
n
th
is
w
or
k,
th
r
e
e
a
s
pe
c
t
s
ha
ve
be
e
n
pr
io
r
it
iz
e
s
w
hi
c
h
a
r
e
pe
r
c
e
pt
ib
il
it
y,
s
e
c
ur
it
y,
a
nd
di
ve
r
s
it
y.
T
o
a
c
hi
e
ve
a
hi
gh
-
qua
li
ty
s
te
go
im
a
ge
,
th
e
y
us
e
d
a
W
G
A
N
in
s
te
a
d of
t
he
c
om
m
onl
y us
e
d D
C
G
A
N
,
w
hi
c
h l
e
a
ds
t
o f
a
s
te
r
t
r
a
in
in
g a
nd s
upe
r
io
r
vi
s
ua
l
qua
li
ty
.
3.
M
E
T
H
O
D
T
he
pr
opos
e
d
m
e
th
od
in
vol
ve
s
de
s
ig
ni
ng
two
a
dva
nc
e
d
a
r
ti
f
ic
ia
l
ne
ur
a
l
ne
twor
ks
(
A
N
N
s
)
a
s
c
or
e
c
om
pone
nt
s
of
th
e
s
y
s
te
m
.
T
he
obj
e
c
ti
ve
is
to
c
r
e
a
te
a
on
e
-
to
-
one
m
a
ppi
ng
be
twe
e
n
two
di
s
ti
nc
t
im
a
ge
s
e
t
s
.
I
de
a
ll
y,
th
e
m
ode
l
s
houl
d
pe
r
f
e
c
tl
y
t
r
a
ns
f
or
m
a
n
in
put
im
a
ge
(
A
)
in
to
a
c
om
pl
e
te
ly
d
if
f
e
r
e
nt
out
put
im
a
ge
(
B
)
.
T
hi
s
im
a
ge
B
c
a
n
be
lo
ng
to
th
e
s
a
m
e
dom
a
in
a
s
im
a
ge
A
or
a
di
f
f
e
r
e
nt
dom
a
in
a
lt
oge
th
e
r
.
C
r
uc
ia
ll
y
,
im
a
ge
B
s
houl
d
e
xhi
bi
t
no
di
s
c
e
r
ni
bl
e
tr
a
c
e
s
of
th
e
in
f
or
m
a
t
io
n
or
ig
in
a
ll
y
c
ont
a
in
e
d
w
it
hi
n
im
a
ge
A
.
T
o
a
c
hi
e
ve
t
hi
s
, t
he
s
ys
te
m
i
nc
or
por
a
te
s
t
w
o g
e
ne
r
a
ti
ve
ne
ur
a
l
ne
t
w
or
ks
:
‒
A
ut
oe
nc
ode
r
:
t
r
a
in
e
d us
in
g t
w
o s
e
ts
of
i
m
a
ge
s
t
o m
a
p t
he
m
one
-
to
-
one
. I
t
is
us
e
d on both t
he
s
e
nde
r
a
nd
r
e
c
e
iv
e
r
s
id
e
s
t
o c
onv
e
r
t
th
e
c
ove
r
i
m
a
ge
(
s
te
go i
m
a
ge
)
t
o t
he
s
e
c
r
e
t
im
a
ge
a
nd vic
e
ve
r
s
a
.
‒
C
yc
le
G
A
N
:
e
nha
nc
e
s
th
e
vi
s
ua
l
qu
a
li
ty
of
th
e
im
a
ge
s
ge
n
e
r
a
te
d
by
th
e
A
ut
oe
nc
ode
r
.
T
he
C
y
c
le
G
A
N
m
ode
l
is
tr
a
in
e
d
to
le
a
r
n
th
e
f
e
a
tu
r
e
s
of
th
e
im
a
ge
s
’
dom
a
in
s
,
e
ns
ur
in
g
hi
gh
-
qua
li
ty
im
a
ge
tr
a
ns
la
ti
on
be
twe
e
n doma
in
s
.
3.1.
D
at
as
e
t
s
T
he
da
ta
s
e
ts
e
m
pl
oye
d
f
or
th
is
s
tu
dy
in
c
lu
de
im
a
ge
s
f
r
om
th
e
C
e
le
bA
a
nd
W
ik
iAr
t
da
ta
s
e
ts
.
T
h
e
s
e
da
ta
s
e
ts
pr
ovi
de
a
di
ve
r
s
e
r
a
nge
of
im
a
ge
s
ne
c
e
s
s
a
r
y
f
or
tr
a
in
in
g
th
e
ge
ne
r
a
ti
ve
m
ode
ls
e
f
f
e
c
ti
ve
ly
.
T
he
C
e
le
bA
da
ta
s
e
t
in
c
lu
de
s
im
a
ge
s
of
c
e
le
br
it
ie
s
,
w
hi
le
th
e
W
ik
iAr
t
da
ta
s
e
t
e
nc
om
pa
s
s
e
s
va
r
io
us
s
ty
le
s
of
Evaluation Warning : The document was created with Spire.PDF for Python.
I
nt
J
A
r
ti
f
I
nt
e
ll
I
S
S
N
:
2252
-
8938
T
r
ans
la
ti
on
-
bas
e
d i
m
age
s
te
ganog
r
aphy
s
y
s
te
m
ut
il
iz
in
g autoe
nc
ode
r
and …
(
T
hak
w
an A
k
r
am
J
aw
ad)
3961
a
r
t
[
22]
,
[
23]
.
T
o
e
va
lu
a
te
th
e
m
od
e
l'
s
pe
r
f
or
m
a
nc
e
in
a
c
o
m
pr
e
he
ns
iv
e
w
a
y,
bot
h
da
ta
s
e
ts
a
r
e
s
pl
it
in
to
tr
a
in
in
g a
nd t
e
s
ti
ng s
e
ts
.
3
.
2
.
A
r
c
h
it
e
c
t
u
r
e
s
3
.
2
.1.
A
u
t
oe
n
c
od
e
r
T
he
id
e
a
of
d
e
s
ig
ni
ng
th
e
e
nc
od
e
r
is
to
r
e
duc
e
th
e
s
pa
ti
a
l
di
m
e
ns
io
ns
of
th
e
in
put
im
a
ge
in
a
gr
a
dua
l
w
a
y
w
hi
le
e
xpa
ndi
ng
th
e
d
e
pt
h
of
th
e
f
e
a
tu
r
e
m
a
ps
a
nd
c
a
pt
u
r
in
g
ba
s
ic
im
a
ge
f
e
a
tu
r
e
s
.
I
n
th
e
b
e
gi
nni
ng
of
th
e
s
tr
uc
tu
r
e
,
th
e
r
e
a
r
e
th
r
e
e
la
ye
r
s
,
a
C
onv2D
la
ye
r
th
a
t
us
e
s
32
f
il
te
r
s
w
it
h
r
e
c
ti
f
ie
d
li
ne
a
r
uni
t
(
R
e
L
U
)
a
c
ti
va
ti
on,
a
M
a
xP
ool
in
g2D
la
ye
r
th
a
t
us
e
s
pool
s
iz
e
of
2
×
2,
a
nd
a
de
ns
e
la
ye
r
c
om
pr
is
in
g
64
un
it
s
w
it
h
R
e
L
U
a
c
ti
va
ti
on.
A
f
te
r
th
a
t,
th
e
pr
oc
e
s
s
w
il
l
be
r
e
pe
a
te
d
w
it
h
C
onv2D
la
ye
r
of
64
f
il
te
r
s
,
a
not
he
r
la
ye
r
of
M
a
xP
ool
in
g2D
,
a
nd
a
de
ns
e
la
ye
r
th
a
t
ha
s
128
uni
ts
.
A
t
th
e
f
in
a
l
s
ta
ge
,
th
e
e
nc
ode
r
ha
s
a
C
onv2D
la
ye
r
th
a
t
ha
ve
128 f
il
te
r
s
, a
M
a
xP
ool
in
g2D
l
a
ye
r
, a
nd a
de
ns
e
l
a
ye
r
t
ha
t
c
ont
a
in
s
256 unit
s
. A
ll
us
e
R
e
L
U
a
c
ti
va
ti
on t
o
gua
r
a
nt
e
e
non
-
li
ne
a
r
it
y a
nd r
obus
t
f
e
a
tu
r
e
e
xt
r
a
c
ti
on.
T
he
f
unc
ti
on
of
th
e
de
c
ode
r
is
to
r
e
f
in
e
th
e
im
a
ge
f
r
om
th
e
e
n
c
ode
d
la
te
nt
il
lu
s
tr
a
ti
on,
th
is
done
by
r
e
ve
r
s
in
g
th
e
di
m
e
ns
io
na
li
ty
r
e
duc
ti
on.
I
t
s
ta
r
ts
w
it
h
th
e
de
ns
e
la
ye
r
th
a
t
ha
s
8
×
8
×
128
uni
ts
,
a
C
onv2DT
r
a
ns
po
s
e
la
ye
r
w
it
h
256
f
il
te
r
s
of
s
iz
e
3
×
3
a
nd a
s
tr
id
e
of
2
a
nd
R
e
L
U
a
c
ti
va
ti
on,
a
nd a
de
ns
e
la
ye
r
w
it
h
64
uni
ts
.
T
he
n,
it
c
ont
in
ue
s
w
it
h
a
C
onv2D
T
r
a
ns
po
s
e
la
ye
r
th
a
t
ha
s
28
f
il
te
r
s
a
nd a
s
tr
id
e
of
2,
f
ol
lo
w
e
d
by
a
de
ns
e
la
ye
r
th
a
t
h
a
s
32
uni
t
s
.
T
h
e
f
in
a
l
s
te
p
h
a
ve
C
onv2D
T
r
a
ns
pos
e
la
ye
r
th
a
t
h
a
s
64
f
il
te
r
s
,
a
s
tr
id
e
of
2,
a
nd
a
c
onc
lu
di
ng
C
onv2D
out
put
la
ye
r
th
a
t
ha
s
w
it
h
3
f
il
te
r
s
of
s
iz
e
3
×
3.
I
t
e
m
pl
oys
s
ig
m
oi
d
a
c
ti
va
ti
on
to
pr
oduc
e
t
he
f
in
a
l
out
put
i
m
a
ge
.
3
.
2
.
2
.
C
yc
le
G
A
N
T
he
a
r
c
hi
te
c
tu
r
e
of
th
e
C
y
c
le
G
A
N
ge
ne
r
a
to
r
in
c
lu
de
s
th
r
e
e
m
a
i
n
m
odul
e
s
w
hi
c
h
a
r
e
dow
n
s
a
m
pl
in
g,
r
e
s
id
ua
l
bl
oc
ks
, a
nd ups
a
m
pl
in
g. T
he
pha
s
e
of
dow
ns
a
m
pl
in
g s
ta
r
ts
w
it
h a
C
onv2D la
ye
r
t
ha
t
ha
s
64 f
il
te
r
s
,
a
4
×
4
ke
r
ne
l,
a
nd
a
s
tr
id
e
of
2.
T
o
r
e
duc
e
th
e
s
pa
ti
a
l
di
m
e
ns
io
ns
by
ha
lf
,
R
e
L
U
a
c
ti
va
ti
on
w
il
l
be
a
dde
d.
T
o
f
ur
th
e
r
ha
lv
e
th
e
di
m
e
ns
io
ns
,
th
e
C
onv2D
la
ye
r
w
il
l
be
us
e
d.
T
hi
s
la
ye
r
ha
s
128
f
il
te
r
s
w
hi
le
f
ol
lo
w
in
g
th
e
s
a
m
e
ke
r
ne
l
a
nd
s
tr
id
e
to
in
c
r
e
a
s
e
f
e
a
tu
r
e
de
pt
h.
T
o
c
a
pt
ur
e
m
or
e
c
om
pl
e
x
f
e
a
tu
r
e
s
w
hi
le
c
ont
in
ui
ng
th
e
s
pa
ti
a
l
di
m
e
ns
io
n
r
e
duc
ti
on,
a
C
onv2D
la
ye
r
w
it
h
256
f
il
te
r
s
,
4
×
4
ke
r
ne
l,
a
nd
s
tr
id
e
of
2
h
a
s
b
e
e
n
us
e
d.
T
o
m
a
in
ta
in
th
e
ne
twor
k'
s
id
e
nt
it
y
m
a
ppi
ng
s
a
nd
tr
a
in
in
g
c
ons
ta
nc
y,
out
s
ta
ndi
ng
bl
oc
ks
w
il
l
be
u
s
e
d a
s
th
e
ir
u
s
e
is
c
r
uc
ia
l.
T
he
s
ix
out
s
ta
ndi
ng
bl
oc
k
s
c
om
pr
is
in
g
a
C
onv2D
l
a
ye
r
w
it
h
256
f
il
te
r
s
,
a
3
×
3
ke
r
ne
l,
a
nd
s
a
m
e
pa
ddi
ng
th
a
t
f
ol
lo
w
e
d
by
R
e
L
U
a
c
ti
va
ti
on.
A
not
he
r
C
onv2D
la
ye
r
th
a
t
ha
s
256
f
il
te
r
s
a
nd
s
a
m
e
pa
ddi
ng
f
ol
lo
w
s
t
ha
t
th
e
i
nput
a
nd output
s
um
m
e
d t
o f
or
m
t
he
r
e
s
id
ua
l
c
onne
c
ti
on, pr
om
ot
in
g s
ta
bl
e
t
r
a
in
in
g
[
24]
.
T
he
ups
a
m
pl
in
g
s
ta
ge
s
ta
r
s
w
it
h
a
C
onv2DT
r
a
n
s
pos
e
la
ye
r
th
a
t
ha
s
256
f
il
te
r
s
,
a
4
×
4
ke
r
ne
l,
a
nd
a
s
tr
id
e
of
2.
I
t
w
il
l
be
c
ti
va
te
d
by
R
e
L
U
,
doubli
ng
th
e
s
pa
ti
a
l
di
m
e
ns
io
ns
,
f
ol
lo
w
e
d
by
a
C
onv2
D
T
r
a
ns
po
s
e
la
ye
r
th
a
t
ha
s
128
f
il
te
r
s
a
nd
th
e
s
a
m
e
k
e
r
ne
l
a
nd
s
tr
id
e
.
T
o
f
ur
th
e
r
ups
a
m
pl
e
s
th
e
di
m
e
ns
io
n
s
,
a
not
he
r
C
onv2DT
r
a
ns
po
s
e
la
ye
r
w
it
h
64
f
il
te
r
s
a
nd
s
im
il
a
r
s
e
tt
in
gs
.
F
i
na
ll
y,
th
e
out
put
la
ye
r
is
a
C
onv2DT
r
a
ns
pos
e
la
ye
r
w
it
h
3
f
il
te
r
s
,
a
4
×
4
ke
r
ne
l,
a
nd
a
s
tr
id
e
of
2.
I
t
w
il
l
be
f
ol
lo
w
e
d
by
S
ig
m
oi
d
a
c
ti
va
ti
on
to
c
ons
tr
a
in
out
put
pi
xe
l
va
lu
e
s
be
twe
e
n 0
a
nd 1
[
25]
, [
26]
.
F
or
th
e
di
s
c
r
im
in
a
to
r
,
it
ha
s
be
e
n
de
s
ig
ne
d
a
s
a
P
a
tc
hG
A
N
th
a
t
f
oc
us
e
s
on
lo
c
a
l
im
a
g
e
pa
tc
h
e
s
r
a
th
e
r
th
a
n
th
e
e
nt
ir
e
im
a
ge
,
m
a
in
ta
in
in
g
hi
gh
-
f
r
e
que
nc
y
de
ta
il
s
a
nd
te
xt
ur
e
s
.
I
t
ha
s
m
ul
ti
pl
e
C
onv2D
la
ye
r
s
w
it
h
f
il
te
r
s
a
nd
s
tr
id
e
s
of
2,
e
a
c
h
f
ol
lo
w
e
d
by
L
e
a
kyR
e
L
U
a
c
ti
va
ti
ons
.
I
t
pr
ogr
e
s
s
iv
e
ly
r
e
duc
in
g
th
e
s
pa
ti
a
l
di
m
e
ns
io
ns
w
hi
le
c
a
pt
ur
in
g
m
or
e
c
om
pl
e
x
f
e
a
tu
r
e
s
. T
he
f
in
a
l
la
ye
r
out
put
s
a
s
in
gl
e
-
c
ha
nne
l
f
e
a
tu
r
e
m
a
p
th
a
t
r
e
pr
e
s
e
nt
s
t
he
a
ut
he
nt
ic
it
y of
e
a
c
h i
m
a
ge
pa
tc
h, pr
om
ot
in
g de
ta
il
e
d a
nd r
e
a
li
s
ti
c
i
m
a
ge
ge
ne
r
a
ti
on
[
9]
, [
27]
.
4.
R
E
S
U
L
T
S
A
N
D
D
I
S
C
U
S
S
I
O
N
T
h
e
f
un
c
t
io
n
a
l
it
y
of
o
ur
pr
op
o
s
e
d m
e
t
ho
d
i
s
e
v
a
lu
a
t
e
d
th
r
ou
gh
v
a
r
io
u
s
pe
r
f
or
m
a
n
c
e
m
e
tr
ic
s
,
i
nc
lu
d
in
g
s
tr
uc
tu
r
a
l
s
i
m
il
a
r
i
ty
in
de
x
(
S
S
I
M
)
[
2
8]
,
m
e
a
n
s
qu
a
r
e
d
e
r
r
or
(
M
S
E
)
,
a
nd
p
e
a
k
s
i
gn
a
l
-
to
-
n
oi
s
e
r
a
ti
o
(
P
S
N
R
)
.
T
o
c
om
pr
e
he
n
s
iv
e
ly
e
va
lu
a
te
our
pr
opo
s
e
d
s
ys
t
e
m
,
w
e
h
a
ve
d
e
f
in
e
d
t
w
o
di
s
ti
n
c
t
s
c
e
na
r
io
s
.
T
h
e
s
e
s
c
e
na
r
io
s
a
r
e
d
e
s
i
gne
d
t
o t
e
s
t
t
he
s
y
s
te
m
'
s
pe
r
f
or
m
a
nc
e
i
n di
f
f
e
r
e
nt
c
ont
e
x
ts
a
nd w
it
h di
f
f
e
r
e
nt
t
y
pe
s
of
i
m
a
ge
s
.
4
.1.
T
r
ai
n
in
g C
yc
le
G
A
N
W
e
s
e
l
e
c
te
d
50
im
a
g
e
s
e
a
c
h
f
r
om
th
e
C
e
le
bA
a
nd
a
c
ti
on
pa
i
nt
in
g
s
ty
le
in
th
e
W
ik
iAr
t
da
ta
s
e
t
to
tr
a
in
th
e
C
yc
le
G
A
N
m
ode
l,
a
im
in
g
to
r
e
pr
e
s
e
nt
two
di
s
ti
nc
t
vi
s
ua
l
dom
a
in
s
a
nd
th
us
pr
ovi
de
a
r
obus
t
te
s
t
f
or
our
m
ode
l'
s
ge
ne
r
a
li
z
a
ti
on
c
a
pa
bi
li
ti
e
s
.
T
he
r
e
s
ul
t
s
a
nd
vi
s
ua
li
z
a
ti
ons
in
c
lu
de
r
e
c
or
de
d
tr
a
in
in
g
lo
s
s
e
s
f
or
bot
h
th
e
ge
ne
r
a
to
r
s
a
nd
di
s
c
r
im
in
a
to
r
s
ove
r
th
e
500
e
poc
hs
.
T
he
pl
ot
in
F
ig
ur
e
1
il
lu
s
tr
a
te
s
th
e
tr
a
in
in
g
dyna
m
ic
s
, s
pe
c
if
ic
a
ll
y t
he
l
os
s
c
ur
ve
s
f
or
bot
h ge
ne
r
a
to
r
s
(
G
e
n
A
to
B
a
nd G
e
n B
to
A
)
a
s
t
he
y l
e
a
r
n t
o t
r
a
ns
la
te
im
a
ge
s
f
r
om
one
dom
a
in
to
a
not
he
r
.
T
h
e
C
yc
le
G
A
N
m
ode
l
w
a
s
tr
a
in
e
d
f
or
500
e
poc
h
s
to
le
a
r
n
th
e
m
a
ppi
ngs
be
twe
e
n
th
e
s
e
two
dom
a
in
s
,
c
a
pt
ur
in
g
a
nd
tr
a
ns
la
ti
ng
th
e
u
ni
que
f
e
a
tu
r
e
s
a
nd
s
ty
le
s
of
e
a
c
h
dom
a
in
,
a
s
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2252
-
8938
I
nt
J
A
r
ti
f
I
nt
e
ll
,
V
ol
. 14, No. 5, O
c
to
be
r
2025
:
3958
-
3969
3962
s
how
n
in
F
ig
ur
e
s
2
,
C
e
le
bA
to
W
ik
iAr
t
a
s
s
how
n
in
F
ig
ur
e
2(
a
)
,
a
nd
W
ik
iAr
t
to
C
e
le
bA
a
s
s
ho
w
n
in
F
ig
ur
e
2(
b)
. T
hi
s
t
r
a
in
in
g e
nha
nc
e
s
t
he
vi
s
u
a
l
im
pe
r
c
e
pt
ib
il
it
y of
t
he
ge
ne
r
a
te
d i
m
a
ge
s
.
F
ig
ur
e
1. G
e
ne
r
a
to
r
s
'
t
r
a
in
in
g l
os
s
ove
r
e
poc
hs
(
a
)
(
b)
F
ig
ur
e
2. E
xa
m
pl
e
s
of
C
yc
le
G
A
N
dom
a
in
-
to
-
dom
a
in
t
r
a
ns
la
ti
on a
f
te
r
t
he
m
ode
l
is
t
r
a
in
e
d
of
(
a
)
C
e
le
bA
t
o
W
ik
iAr
t
a
nd
(
b)
W
ik
iAr
t
to
C
e
le
bA
4
.2.
S
c
e
n
ar
io
1:
s
am
e
d
o
m
ai
n
I
n
th
is
s
c
e
na
r
io
,
bot
h
th
e
s
e
c
r
e
t
im
a
ge
a
nd
th
e
s
te
go
im
a
ge
be
lo
ng
to
th
e
s
a
m
e
dom
a
in
,
s
ha
r
in
g
s
im
il
a
r
c
ha
r
a
c
te
r
is
ti
c
s
s
uc
h
a
s
s
ty
le
,
c
ont
e
nt
,
a
nd
vi
s
ua
l
f
e
a
tu
r
e
s
.
E
va
lu
a
ti
ng
th
e
s
ys
te
m
unde
r
th
is
c
ondi
ti
o
n
he
lp
s
us
unde
r
s
ta
nd
it
s
pe
r
f
or
m
a
nc
e
w
he
n
th
e
im
a
ge
s
ha
ve
a
hi
gh
de
gr
e
e
of
s
im
il
a
r
it
y.
W
e
s
e
le
c
te
d
20
im
a
ge
s
f
r
om
th
e
C
e
le
bA
da
ta
s
e
t,
f
e
a
tu
r
in
g
bot
h
m
a
le
a
nd
f
e
m
a
le
s
ubj
e
c
ts
w
it
h
di
ve
r
s
e
ba
c
kgr
ounds
,
to
te
s
t
th
e
s
ys
te
m
'
s
a
bi
li
ty
to
m
a
p
a
nd
tr
a
ns
la
te
f
a
c
ia
l
im
a
ge
s
.
T
h
e
s
e
im
a
ge
s
,
w
it
h
r
e
la
ti
ve
ly
s
im
pl
e
r
a
nd
m
or
e
uni
f
or
m
vi
s
ua
l
f
e
a
tu
r
e
s
c
om
pa
r
e
d
to
a
r
ti
s
ti
c
im
a
g
e
s
,
pr
ovi
de
a
n
a
ppr
opr
ia
te
te
s
t
c
a
s
e
to
a
s
s
e
s
s
th
e
s
y
s
te
m
'
s
e
f
f
ic
a
c
y
in
ha
ndl
in
g
f
a
c
ia
l
im
a
ge
r
y.
I
n
th
e
in
it
ia
l
ph
a
s
e
of
our
e
va
lu
a
ti
on,
w
e
f
oc
us
e
d
on
th
e
f
ir
s
t
s
c
e
na
r
io
,
w
he
r
e
bot
h
th
e
s
te
go
-
im
a
ge
a
nd
th
e
s
e
c
r
e
t
im
a
ge
or
ig
in
a
t
e
f
r
om
th
e
s
a
m
e
dom
a
in
.
T
hi
s
s
c
e
na
r
io
is
pa
r
ti
c
ul
a
r
ly
s
ig
ni
f
ic
a
nt
a
s
it
a
ll
ow
s
u
s
to
a
s
s
e
s
s
th
e
pe
r
f
or
m
a
nc
e
of
our
m
ode
l
unde
r
c
ondi
ti
ons
w
he
r
e
th
e
f
e
a
tu
r
e
s
of
t
he
i
m
a
ge
s
a
r
e
hi
ghl
y s
im
il
a
r
.
T
he
s
im
il
a
r
it
y
in
f
e
a
tu
r
e
s
be
twe
e
n
th
e
s
te
go
-
im
a
ge
a
nd
th
e
s
e
c
r
e
t
im
a
ge
s
im
pl
if
ie
s
th
e
le
a
r
ni
ng
pr
oc
e
s
s
f
or
our
m
ode
l,
th
e
r
e
by
s
e
r
vi
ng
a
s
a
b
a
s
e
li
ne
f
or
e
va
lu
a
ti
ng
it
s
e
f
f
e
c
ti
ve
ne
s
s
.
I
n
th
is
e
va
lu
a
ti
on,
w
e
us
e
d
20
im
a
ge
s
f
r
om
th
e
C
e
le
bA
d
a
ta
s
e
t.
T
he
s
e
im
a
ge
s
f
e
a
tu
r
e
d
bot
h
m
a
le
a
nd
f
e
m
a
le
s
ubj
e
c
t
s
w
it
h
va
r
ie
d
ba
c
kgr
ounds
.
W
e
r
a
ndoml
y
s
e
le
c
te
d
20
im
a
ge
s
a
nd
s
ubs
e
que
nt
ly
di
vi
de
d
th
e
m
in
to
two
di
s
ti
nc
t
s
e
ts
,
la
be
le
d
A
a
nd
B
,
a
s
il
lu
s
tr
a
te
d
in
th
e
F
ig
ur
e
3
.
T
hi
s
pr
oc
e
s
s
e
n
s
ur
e
s
a
ba
la
nc
e
d
a
nd
r
e
pr
e
s
e
nt
a
ti
ve
s
a
m
pl
e
,
w
hi
c
h
is
Evaluation Warning : The document was created with Spire.PDF for Python.
I
nt
J
A
r
ti
f
I
nt
e
ll
I
S
S
N
:
2252
-
8938
T
r
ans
la
ti
on
-
bas
e
d i
m
age
s
te
ganog
r
aphy
s
y
s
te
m
ut
il
iz
in
g autoe
nc
ode
r
and …
(
T
hak
w
an A
k
r
am
J
aw
ad)
3963
e
s
s
e
nt
ia
l
f
or
e
va
lu
a
ti
on
of
our
pr
opos
e
d
m
e
th
od.
F
ig
ur
e
4
il
lu
s
tr
a
te
s
th
e
a
ut
oe
nc
ode
r
'
s
tr
a
in
in
g
lo
s
s
tr
a
je
c
to
r
y
ove
r
2000
e
poc
hs
,
r
e
f
le
c
ti
ng
th
e
opt
im
iz
a
ti
on
pr
oc
e
s
s
.
F
ol
l
ow
in
g
th
is
,
F
ig
ur
e
5
e
va
lu
a
te
s
th
e
m
ode
l'
s
pe
r
f
or
m
a
nc
e
by
s
how
in
g
or
ig
in
a
l
im
a
ge
s
in
F
ig
ur
e
5
(
a
)
,
th
e
ir
ge
ne
r
a
te
d
c
ount
e
r
pa
r
ts
in
F
ig
ur
e
5(
b)
,
a
nd
th
e
ta
r
ge
t
s
e
c
r
e
t
im
a
ge
s
i
n
F
ig
ur
e
5(
c
)
, hi
ghl
ig
ht
in
g t
he
m
ode
l'
s
r
e
c
ons
tr
uc
ti
on a
c
c
ur
a
c
y.
F
ig
ur
e
3. T
he
im
a
ge
s
of
t
w
o s
e
t
s
t
ha
t
w
il
l
be
m
a
ppe
d c
or
r
e
s
pon
di
ngl
y
F
ig
ur
e
4. T
he
t
r
a
in
in
g l
os
s
ove
r
e
poc
h
s
dur
in
g t
he
t
r
a
in
in
g of
t
h
e
a
ut
oe
nc
ode
r
on t
he
C
e
l
e
bA
da
ta
s
e
t
(
a
)
(
b)
(
c
)
F
ig
ur
e
5. A
ut
oe
nc
ode
r
m
ode
l
e
va
lu
a
ti
on
of
(
a
)
in
put
:
or
ig
in
a
l
i
m
a
ge
,
(
b)
out
put
:
ge
ne
r
a
te
d i
m
a
ge
,
a
nd
(
c
)
ta
r
ge
t:
s
e
c
r
e
t
im
a
ge
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2252
-
8938
I
nt
J
A
r
ti
f
I
nt
e
ll
,
V
ol
. 14, No. 5, O
c
to
be
r
2025
:
3958
-
3969
3964
A
s
s
how
n
in
T
a
bl
e
1,
t
he
a
na
ly
s
is
of
th
e
C
e
le
bA
-
to
-
C
e
le
bA
tr
a
ns
la
ti
on
pe
r
f
or
m
a
nc
e
f
or
th
e
a
ut
oe
nc
ode
r
m
ode
l
r
e
ve
a
ls
not
e
w
or
th
y
f
in
di
ngs
f
r
om
th
e
c
om
p
a
r
a
ti
ve
a
s
s
e
s
s
m
e
nt
of
S
S
I
M
,
M
S
E
,
a
nd
P
S
N
R
m
e
tr
ic
s
a
c
r
os
s
di
f
f
e
r
e
nt
im
a
ge
s
e
ts
.
I
n
th
e
A
to
B
m
od
e
l,
th
e
r
e
s
ul
ts
de
m
ons
tr
a
te
c
on
s
is
te
nt
ly
hi
gh
S
S
I
M
va
lu
e
s
,
pe
a
ki
ng
a
t
0.908185
f
or
im
a
ge
s
e
t
4,
in
di
c
a
ti
ng
s
tr
o
ng
pr
e
s
e
r
va
ti
on
of
s
tr
uc
tu
r
a
l
in
te
gr
it
y
dur
in
g
tr
a
ns
la
ti
on.
A
ddi
ti
ona
ll
y,
th
is
m
ode
l
e
xhi
bi
ts
lo
w
M
S
E
va
lu
e
s
,
pa
r
ti
c
ul
a
r
ly
not
a
bl
e
in
i
m
a
ge
s
e
t
4
w
it
h
a
n
M
S
E
of
0.000360,
a
nd
a
c
or
r
e
s
ponding
P
S
N
R
of
34.434243,
s
ig
ni
f
yi
ng
m
in
im
a
l
r
e
c
ons
tr
uc
ti
on
e
r
r
or
a
nd
hi
gh
-
qua
li
ty
out
put
im
a
ge
s
.
T
he
s
e
r
e
s
ul
ts
c
ol
le
c
ti
ve
ly
hi
ghl
ig
ht
th
e
r
obus
tn
e
s
s
of
th
e
a
ut
oe
n
c
ode
r
m
ode
l
in
m
a
in
ta
in
in
g
im
a
ge
qua
li
ty
,
w
hi
le
a
ls
o
s
ugge
s
ti
ng
a
r
e
a
s
f
or
im
pr
ove
m
e
nt
in
a
c
hi
e
vi
ng
c
ons
is
te
nt
pe
r
f
or
m
a
nc
e
a
c
r
os
s
di
f
f
e
r
e
nt
i
m
a
ge
s
e
t
s
.
T
a
bl
e
1. E
va
lu
a
ti
on me
tr
ic
s
f
or
ge
ne
r
a
te
d i
m
a
ge
s
i
n
F
ig
ur
e
5
I
m
a
ge
s
e
t
S
S
I
M
M
S
E
P
S
N
R
I
m
a
ge
s
e
t
1
0.859619
0.003551
24.496793
I
m
a
ge
s
e
t
2
0.825956
0.000952
30.212155
I
m
a
ge
s
e
t
3
0.866684
0.002285
26.41
101
1
I
m
a
ge
s
e
t
4
0.908185
0.000360
34.434243
I
m
a
ge
s
e
t
5
0.864702
0.000943
30.256875
I
m
a
ge
s
e
t
6
0.855163
0.000631
32.002772
I
m
a
ge
s
e
t
7
0.7591
18
0.001630
27.877301
I
m
a
ge
s
e
t
8
0.904223
0.000509
32.932824
I
m
a
ge
s
e
t
9
0.895943
0.000760
31.191
168
I
m
a
ge
s
e
t
10
0.863341
0.000718
31.439247
4
.
3
.
S
c
e
n
ar
io
2:
d
if
f
e
r
e
n
t
d
om
ai
n
s
T
he
s
e
c
r
e
t
im
a
ge
a
nd
th
e
s
te
go
im
a
ge
c
om
e
f
r
om
di
f
f
e
r
e
nt
d
om
a
in
s
in
th
is
s
c
e
na
r
io
to
pr
e
s
e
nt
a
m
or
e
c
ha
ll
e
ngi
ng
s
e
tu
p
due
to
th
e
ir
di
s
s
im
il
a
r
c
h
a
r
a
c
te
r
is
ti
c
s
in
s
ty
le
,
c
ont
e
nt
,
a
nd
vi
s
ua
l
f
e
a
tu
r
e
s
.
T
he
e
va
lu
a
ti
on
od
th
is
s
ys
te
m
unde
r
th
e
s
e
c
ondi
ti
ons
he
lp
s
to
un
de
r
s
ta
nd
it
s
r
obus
tn
e
s
s
a
nd
ve
r
s
a
ti
li
ty
.
S
e
c
r
e
t
im
a
ge
s
f
r
om
th
e
C
e
le
bA
da
ta
s
e
t
a
nd
s
te
go
im
a
ge
s
f
r
om
th
e
W
ik
iAr
t
da
ta
s
e
t
ha
v
e
be
e
n
us
e
d
to
t
e
s
t
th
e
s
ys
te
m
'
s
a
bi
li
ty
to
m
a
p
a
nd
tr
a
ns
la
te
f
a
c
ia
l
im
a
ge
s
to
a
r
ti
s
ti
c
i
m
a
ge
s
.
T
he
obj
e
c
ti
ve
w
a
s
e
ns
ur
in
g
th
e
f
a
c
ia
l
f
e
a
tu
r
e
s
w
e
r
e
pr
e
s
e
r
v
in
g
w
hi
le
m
a
in
ta
in
in
g
th
e
a
r
ti
s
ti
c
in
te
gr
it
y
of
th
e
s
te
go
im
a
ge
s
.
I
n
th
e
s
e
c
ond
pha
s
e
of
th
e
e
va
lu
a
ti
on
,
it
h
a
s
b
e
e
n
f
oc
u
s
e
d
on
th
e
s
e
c
ond
s
c
e
na
r
io
,
w
h
e
r
e
th
e
s
t
e
go
-
im
a
ge
a
nd
th
e
s
e
c
r
e
t
im
a
ge
ha
ve
be
e
n
or
ig
in
a
te
d
f
r
om
di
f
f
e
r
e
nt
dom
a
in
s
.
T
hi
s
s
c
e
na
r
io
i
s
c
r
uc
ia
l
a
s
it
gi
ve
s
th
e
a
bi
li
ty
to
e
va
lu
a
te
th
e
pe
r
f
or
m
a
nc
e
of
th
e
m
ode
l
unde
r
c
ondi
ti
ons
w
he
r
e
th
e
f
e
a
tu
r
e
s
of
th
e
im
a
ge
s
a
r
e
hi
ghl
y
di
f
f
e
r
e
nt
a
nd
uns
im
il
a
r
.
I
t
is
im
por
ta
nt
to
m
e
nt
io
n
th
a
t
th
e
di
f
f
e
r
e
nc
e
in
f
e
a
tu
r
e
s
be
twe
e
n
th
e
s
te
go
-
im
a
ge
a
nd
th
e
s
e
c
r
e
t
im
a
ge
c
om
pl
ic
a
te
s
t
he
l
e
a
r
ni
ng pr
oc
e
s
s
f
or
our
m
ode
l.
A
s
s
how
n
in
F
ig
ur
e
6
,
twe
nt
y
im
a
ge
s
w
e
r
e
r
a
ndoml
y
s
e
le
c
te
d
f
or
e
va
lu
a
ti
on
.
T
he
s
e
im
a
g
e
s
ha
v
e
be
e
n
s
e
le
c
te
d
f
r
om
th
e
W
ik
iAr
t
a
nd
C
e
le
bA
d
a
ta
s
e
t
s
e
qu
a
ll
y.
T
he
r
e
a
s
on
f
or
c
hoos
in
g
th
e
s
e
im
a
ge
s
i
s
to
c
ove
r
a
di
ve
r
s
e
r
a
nge
of
a
r
ti
s
ti
c
s
ty
le
s
a
nd
in
tr
ic
a
te
vi
s
ua
l
a
nd
f
a
c
ia
l
f
e
a
tu
r
e
s
to
a
s
s
e
s
s
th
e
a
bi
li
ty
of
th
e
s
ys
te
m
to
t
r
a
ns
la
te
a
nd ma
p i
m
a
ge
s
a
c
r
os
s
di
f
f
e
r
e
nt
doma
in
s
.
I
n
F
ig
ur
e
7, t
he
a
ut
oe
nc
ode
r
'
s
t
r
a
in
in
g
l
os
s
ove
r
e
poc
hs
f
or
bot
h
d
a
ta
s
e
ts
,
th
is
hi
ghl
ig
ht
s
th
e
opt
im
iz
a
ti
on
p
r
oc
e
s
s
a
nd
th
e
m
ode
l'
s
c
onve
r
ge
nc
e
dur
in
g
tr
a
in
in
g
. W
hi
le
F
ig
ur
e
8
i
s
us
e
d t
o
e
va
lu
a
te
t
he
a
ut
oe
nc
ode
r
'
s
pe
r
f
or
m
a
nc
e
by s
how
c
a
s
in
g t
he
or
ig
in
a
l
im
a
ge
s
in
F
ig
ur
e
8(
a
)
,
th
e
n
th
e
ge
ne
r
a
te
d
im
a
ge
s
in
F
ig
ur
e
8(
b)
,
a
nd
th
e
ta
r
ge
t
s
e
c
r
e
t
im
a
ge
s
in
F
ig
ur
e
8(
c
)
,
de
m
ons
tr
a
ti
ng t
he
m
ode
l'
s
c
a
pa
c
it
y f
or
e
f
f
e
c
ti
ve
r
e
c
ons
tr
uc
ti
on a
nd doma
in
m
a
ppi
ng.
F
ig
ur
e
6.
T
he
im
a
ge
s
of
t
w
o s
e
t
s
t
ha
t
w
il
l
be
m
a
ppe
d c
or
r
e
s
pon
di
ngl
y
Evaluation Warning : The document was created with Spire.PDF for Python.
I
nt
J
A
r
ti
f
I
nt
e
ll
I
S
S
N
:
2252
-
8938
T
r
ans
la
ti
on
-
bas
e
d i
m
age
s
te
ganog
r
aphy
s
y
s
te
m
ut
il
iz
in
g autoe
nc
ode
r
and …
(
T
hak
w
an A
k
r
am
J
aw
ad)
3965
F
ig
ur
e
7.
T
h
e
t
r
a
i
ni
n
g l
o
s
s
o
ve
r
e
po
c
h
s
dur
in
g t
he
t
r
a
in
in
g
of
t
h
e
a
ut
oe
nc
od
e
r
on
t
h
e
W
i
ki
A
r
t/
C
e
le
bA
d
a
t
a
s
e
t
s
(
a
)
(
b)
(
c
)
F
ig
ur
e
8. A
ut
oe
nc
ode
r
m
ode
l
e
va
lu
a
ti
on
of
(
a
)
in
put
:
or
ig
in
a
l
i
m
a
ge
,
(
b)
out
put
:
ge
ne
r
a
te
d i
m
a
ge
,
a
nd
(
c
)
ta
r
ge
t:
s
e
c
r
e
t
im
a
ge
I
n
a
ddi
ti
on,
T
a
bl
e
2
c
ont
r
ib
ut
e
s
a
qua
nt
it
a
ti
ve
a
s
s
e
s
s
m
e
nt
o
f
th
e
a
ut
oe
nc
ode
r
'
s
pe
r
f
or
m
a
nc
e
.
I
t
pr
e
s
e
nt
s
e
va
lu
a
ti
on
m
e
tr
ic
s
f
or
th
e
ge
ne
r
a
te
d
im
a
ge
s
di
s
pl
a
ye
d
in
F
ig
ur
e
8.
T
he
s
e
m
e
tr
ic
s
c
onf
ir
m
th
e
e
f
f
e
c
ti
ve
ne
s
s
of
t
he
m
ode
l
to
m
a
in
ta
i
n
hi
gh
a
c
c
ur
a
c
y
a
nd qua
li
t
y i
n t
he
r
e
c
ons
tr
uc
te
d i
m
a
ge
s
.
T
a
bl
e
2. E
va
lu
a
ti
on me
tr
ic
s
f
or
ge
ne
r
a
te
d i
m
a
ge
s
in
F
ig
ur
e
8
I
m
a
ge
s
e
t
S
S
I
M
M
S
E
P
S
N
R
I
m
a
ge
s
e
t
1
0.596109
0.009241
20.342598
I
m
a
ge
s
e
t
2
0.649340
0.007551
21.219742
I
m
a
ge
s
e
t
3
0.493413
0.005901
22.290604
I
m
a
ge
s
e
t
4
0.697044
0.005046
22.970548
I
m
a
ge
s
e
t
5
0.628170
0.004263
23.703005
I
m
a
ge
s
e
t
6
0.719750
0.004531
23.437798
I
m
a
ge
s
e
t
7
0.620793
0.006865
21.633631
I
m
a
ge
s
e
t
8
0.673014
0.00731
1
21.360182
I
m
a
ge
s
e
t
9
0.707038
0.00721
1
21.419772
I
m
a
ge
s
e
t
10
0.476001
0.00741
1
21.301239
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2252
-
8938
I
nt
J
A
r
ti
f
I
nt
e
ll
,
V
ol
. 14, No. 5, O
c
to
be
r
2025
:
3958
-
3969
3966
4
.3.1. Ap
p
ly
in
g C
yc
le
G
A
N
A
s
th
e
pr
e
vi
ous
r
e
s
ul
ts
d
e
m
ons
tr
a
te
d,
th
e
a
ut
oe
n
c
ode
r
'
s
pe
r
f
or
m
a
nc
e
on
im
a
ge
pa
ir
s
1,
3,
a
nd
10
w
a
s
n'
t
good
e
nough.
T
he
r
e
f
or
e
,
in
th
is
s
e
c
ti
on,
w
e
a
r
e
goi
ng
to
a
ppl
y
our
tr
a
in
e
d
C
yc
le
G
A
N
to
th
e
im
a
ge
s
th
a
t
w
e
r
e
out
put
te
d
f
r
om
th
e
a
ut
oe
nc
ode
r
to
e
nha
nc
e
th
e
m
.
W
e
w
il
l
th
e
n
c
om
pa
r
e
th
e
s
e
e
nha
n
c
e
d
im
a
ge
s
bot
h
m
e
tr
ic
a
ll
y
a
nd
vi
s
u
a
ll
y
to
th
e
ir
r
e
s
p
e
c
ti
ve
ta
r
ge
t
im
a
ge
s
.
F
ig
ur
e
9
e
v
a
lu
a
te
s
th
e
C
y
c
le
G
A
N
m
ode
l,
s
how
c
a
s
in
g
it
s
c
a
pa
bi
li
ty
to
e
nha
nc
e
th
e
vi
s
ua
l
qua
li
ty
of
th
e
a
ut
oe
nc
ode
r
'
s
out
put
.
F
ig
ur
e
9(
a
)
di
s
pl
a
ys
th
e
or
ig
in
a
l
im
a
ge
,
F
ig
u
r
e
9(
b
)
pr
e
s
e
nt
s
th
e
im
pr
ove
d
ou
tp
ut
g
e
ne
r
a
te
d
by
th
e
C
yc
le
G
A
N
,
a
nd
F
ig
ur
e
9(
c
)
il
lu
s
tr
a
te
s
th
e
ta
r
ge
t
s
e
c
r
e
t
im
a
ge
,
e
m
pha
s
iz
in
g
th
e
m
ode
l'
s
c
o
nt
r
ib
ut
io
n
to
r
e
f
in
in
g
out
pu
t
qua
li
ty
.
T
a
bl
e
3
pr
ovi
de
s
a
c
om
pa
r
a
ti
ve
a
na
ly
s
is
of
e
va
lu
a
ti
on
m
e
tr
ic
s
f
or
bot
h
th
e
ge
ne
r
a
te
d
im
a
ge
s
f
r
om
th
e
a
ut
oe
nc
ode
r
a
nd
th
e
e
nha
nc
e
d
im
a
ge
s
pr
oduc
e
d
by
th
e
C
yc
le
G
A
N
.
T
he
s
e
m
e
tr
ic
s
il
lu
s
tr
a
te
th
e
im
p
r
ove
m
e
nt
s
in
vi
s
ua
l
qua
li
ty
a
nd f
id
e
li
ty
.
(
a
)
(
b)
(
c
)
F
ig
ur
e
9. C
yc
le
G
A
N
m
ode
l
e
va
lu
a
ti
on
of
(
a
)
in
put
:
or
ig
in
a
l
im
a
ge
,
(
b)
out
put
:
ge
ne
r
a
te
d i
m
a
ge
,
a
nd
(
c
)
ta
r
ge
t:
s
e
c
r
e
t
im
a
ge
T
a
bl
e
3. E
va
lu
a
ti
on me
tr
ic
s
f
or
t
he
ge
ne
r
a
te
d i
m
a
ge
s
a
nd e
nha
n
c
e
d i
m
a
ge
s
I
m
a
ge
s
e
t
M
ode
l
S
S
I
M
M
S
E
P
S
N
R
1
A
ut
oe
nc
ode
r
0.596109
0.009241
20.342598
C
yc
l
e
G
A
N
0.604278
0.009573
20.189400
3
A
ut
oe
nc
ode
r
0.493413
0.005901
22.290604
C
yc
l
e
G
A
N
0.505235
0.005854
22.325549
10
A
ut
oe
nc
ode
r
0.476001
0.007411
21.301239
C
yc
l
e
G
A
N
0.481077
0.007629
21.175290
4
.4.
C
om
p
ar
is
on
o
f
s
t
e
gan
ogr
ap
h
y t
e
c
h
n
iq
u
e
s
T
o
e
va
lu
a
te
th
e
e
f
f
e
c
ti
ve
ne
s
s
of
va
r
io
us
s
te
g
a
nogr
a
phi
c
te
c
h
ni
que
s
,
w
e
c
onduc
te
d
a
c
om
p
a
r
a
ti
ve
a
na
ly
s
is
ba
s
e
d
on
ke
y
f
a
c
to
r
s
:
c
ove
r
im
a
ge
s
iz
e
,
s
e
c
r
e
t
im
a
ge
s
iz
e
,
s
e
c
ur
it
y,
vi
s
ua
l
f
id
e
li
ty
,
a
nd
ve
r
s
a
ti
li
ty
.
T
he
c
ov
e
r
/s
te
go
im
a
ge
s
iz
e
r
e
f
e
r
s
to
th
e
r
e
s
ol
ut
io
n
of
th
e
i
m
a
ge
us
e
d
f
or
e
m
be
ddi
ng
th
e
s
e
c
r
e
t
da
ta
in
tr
a
di
ti
ona
l
s
te
ga
nogr
a
phi
c
m
e
th
ods
a
nd
a
ls
o
r
e
pr
e
s
e
nt
s
th
e
r
e
s
ol
ut
io
n
of
th
e
tr
a
ns
m
it
te
d
im
a
ge
r
e
qui
r
e
d
to
e
xt
r
a
c
t
th
e
hi
dde
n
da
ta
.
S
e
c
r
e
t
im
a
ge
s
iz
e
d
e
not
e
s
th
e
r
e
s
ol
ut
io
n
of
th
e
da
ta
be
in
g
c
onc
e
a
le
d
w
it
hi
n
th
e
c
ove
r
Evaluation Warning : The document was created with Spire.PDF for Python.
I
nt
J
A
r
ti
f
I
nt
e
ll
I
S
S
N
:
2252
-
8938
T
r
ans
la
ti
on
-
bas
e
d i
m
age
s
te
ganog
r
aphy
s
y
s
te
m
ut
il
iz
in
g autoe
nc
ode
r
and …
(
T
hak
w
an A
k
r
am
J
aw
ad)
3967
im
a
ge
.
S
e
c
ur
it
y
m
e
a
s
ur
e
s
th
e
r
e
s
is
t
a
nc
e
of
th
e
ge
ne
r
a
te
d
im
a
ge
s
a
ga
in
s
t
d
e
te
c
ti
on
by
s
te
ga
n
a
ly
s
is
to
ol
s
,
e
ns
ur
in
g
th
a
t
th
e
hi
dde
n
da
ta
r
e
m
a
in
s
im
pe
r
c
e
pt
ib
le
.
V
e
r
s
a
ti
li
ty
a
s
s
e
s
s
e
s
th
e
m
ode
l’
s
a
da
pt
a
bi
li
ty
to
di
f
f
e
r
e
nt
im
a
ge
dom
a
in
s
,
w
hi
le
vi
s
ua
l
f
id
e
li
ty
r
e
f
le
c
ts
th
e
qua
li
ty
a
nd
r
e
a
li
s
m
of
th
e
ge
ne
r
a
t
e
d
im
a
ge
s
.
I
n
T
a
bl
e
4,
w
e
pr
ovi
de
a
de
ta
il
e
d c
om
pa
r
is
on of
our
m
e
th
od w
it
h e
xi
s
ti
ng t
e
c
h
ni
que
s
.
T
a
bl
e
4. C
om
pa
r
is
on b
e
twe
e
n our
pr
opos
e
d m
e
th
od a
nd pr
e
vi
o
us
m
e
th
ods
M
e
t
hod
T
e
c
hni
que
C
ove
r
/
s
t
e
go
i
m
a
ge
s
i
z
e
S
e
c
r
e
t
i
m
a
ge
s
i
z
e
S
e
c
ur
i
t
y
V
i
s
ua
l
f
i
de
l
i
t
y
V
e
r
s
a
t
i
l
i
t
y
P
r
opos
e
d
m
e
t
hod
A
ut
oe
nc
ode
r
+C
yc
l
e
G
A
N
256×256×
3
256×256×
3
H
i
gh
M
ode
r
a
t
e
V
e
r
y
hi
gh
R
e
hm
a
n
’
s
m
e
t
hod
[
29]
E
nc
ode
r
-
d
e
c
ode
r
32×32×
3
32×32×
1
M
ode
r
a
t
e
H
i
gh
H
i
gh
Z
ha
ng’
s
m
e
t
hod
[
30]
GAN
256×256×
3
256×256×
1
M
ode
r
a
t
e
H
i
gh
H
i
gh
4
.5.
L
im
it
at
io
n
s
I
n
th
is
s
tu
dy
,
o
ne
s
i
gn
i
f
ic
a
n
t
l
im
i
ta
t
io
n
is
t
he
s
c
a
r
c
it
y
of
e
s
ta
bl
is
he
d
ba
s
e
l
i
ne
m
od
e
l
s
f
o
r
c
o
m
p
a
r
a
ti
ve
e
v
a
l
ua
t
io
n.
S
i
nc
e
t
hi
s
f
ie
l
d
i
s
r
e
la
ti
ve
l
y
n
a
s
c
e
nt
,
t
he
r
e
is
a
n
ot
ic
e
a
bl
e
l
a
c
k
o
f
s
ta
n
da
r
di
z
e
d
d
a
ta
s
e
ts
a
n
d
p
e
r
f
o
r
m
a
nc
e
m
e
t
r
ic
s
th
a
t
c
a
n
s
e
r
v
e
a
s
b
e
n
c
hm
a
r
ks
f
o
r
a
s
s
e
s
s
i
ng
th
e
e
f
f
e
c
t
iv
e
n
e
s
s
o
f
ou
r
m
o
de
l
.
T
h
is
g
a
p
p
r
e
s
e
n
ts
c
h
a
l
l
e
n
ge
s
i
n
qu
a
n
t
it
a
t
i
ve
ly
e
va
lu
a
t
i
n
g
o
u
r
a
p
p
r
o
a
c
h
a
ga
in
s
t
a
di
ve
r
s
e
a
r
r
a
y
o
f
m
e
th
o
ds
.
M
o
vi
n
g
f
o
r
w
a
r
d
,
th
e
e
s
ta
bl
is
h
m
e
n
t
o
f
s
t
a
n
da
r
d
iz
e
d
b
e
nc
hm
a
r
ks
a
n
d
e
va
lu
a
t
i
on
f
r
a
m
e
w
or
ks
w
i
l
l
b
e
c
r
i
t
ic
a
l
i
n
f
a
c
i
li
ta
t
in
g
m
o
r
e
r
o
bus
t
c
om
pa
r
is
ons
a
nd
f
o
s
t
e
r
i
n
g
a
d
va
nc
e
m
e
n
ts
a
c
r
os
s
v
a
r
i
ou
s
m
o
de
ls
i
n
t
h
is
d
o
m
a
in
.
A
dd
i
ti
on
a
ll
y,
w
h
il
e
t
he
C
y
c
l
e
G
A
N
e
f
f
e
c
t
i
ve
ly
e
n
ha
nc
e
s
th
e
vi
s
u
a
l
q
ua
l
it
y
o
f
t
he
g
e
n
e
r
a
te
d
i
m
a
g
e
s
,
i
t
d
oe
s
no
t
a
c
h
ie
ve
c
o
m
p
le
te
e
n
ha
nc
e
m
e
n
t,
le
a
v
i
ng
s
o
m
e
r
oo
m
f
o
r
im
p
r
ove
m
e
n
t
i
n
te
r
m
s
o
f
i
m
a
g
e
f
i
de
l
it
y
.
F
u
tu
r
e
w
o
r
k
c
o
u
ld
e
x
p
lo
r
e
o
pt
i
m
i
z
in
g
t
he
t
r
a
in
i
ng
p
r
o
c
e
s
s
f
o
r
th
e
a
ut
oe
nc
od
e
r
a
nd
i
m
p
r
ov
in
g
t
he
C
yc
le
G
A
N
'
s
a
r
c
hi
te
c
t
u
r
e
to
a
c
h
ie
ve
h
i
gh
e
r
l
e
ve
ls
o
f
e
nh
a
n
c
e
m
e
nt
,
t
he
r
e
b
y
a
d
d
r
e
s
s
i
ng
t
h
e
s
e
c
ha
l
le
n
ge
s
.
5.
C
O
N
C
L
U
S
I
O
N
T
hi
s
s
tu
dy
de
m
ons
tr
a
te
s
th
e
f
e
a
s
ib
il
it
y
of
us
in
g
ge
ne
r
a
ti
ve
m
ode
ls
,
pa
r
ti
c
ul
a
r
ly
C
yc
le
G
A
N
a
nd
a
ut
oe
nc
ode
r
s
,
f
or
c
ove
r
le
s
s
im
a
ge
s
te
ga
nogr
a
phy,
a
c
hi
e
vi
ng
s
ig
ni
f
ic
a
nt
a
dva
nc
e
m
e
nt
s
in
s
e
c
ur
it
y,
da
ta
c
a
pa
c
it
y,
a
nd
vi
s
ua
l
qua
li
ty
.
U
nl
ik
e
tr
a
di
ti
ona
l
m
e
th
ods
th
a
t
m
odi
f
y
c
a
r
r
ie
r
im
a
ge
s
,
our
a
ppr
oa
c
h
ge
ne
r
a
te
s
unde
te
c
ta
bl
e
s
te
go
im
a
g
e
s
.
T
he
in
te
gr
a
ti
on
of
C
yc
l
e
G
A
N
e
ns
ur
e
s
hi
gh
vi
s
u
a
l
f
id
e
li
ty
,
w
hi
le
a
ut
oe
nc
ode
r
s
e
na
bl
e
e
f
f
e
c
ti
ve
da
ta
c
onc
e
a
lm
e
nt
a
nd
r
e
c
ove
r
y.
T
he
s
e
f
in
di
ngs
hi
ghl
ig
ht
th
e
r
obu
s
tn
e
s
s
,
e
f
f
ic
ie
nc
y,
a
nd
a
ppl
ic
a
bi
li
ty
of
th
e
pr
opos
e
d
m
e
th
od
a
c
r
os
s
v
a
r
io
us
dom
a
in
s
,
pa
vi
ng
th
e
w
a
y
f
or
m
or
e
s
e
c
ur
e
a
nd
pr
a
c
ti
c
a
l
s
ol
ut
io
ns
in
s
te
ga
nogr
a
phy.
F
ut
ur
e
r
e
s
e
a
r
c
h
w
il
l
f
oc
us
on
ove
r
c
om
in
g
li
m
it
a
ti
ons
a
nd
f
u
r
th
e
r
opt
im
iz
in
g
th
e
m
e
th
od f
or
r
e
a
l
-
w
or
ld
a
ppl
ic
a
ti
ons
.
F
U
N
D
I
N
G
I
N
F
O
R
M
A
T
I
O
N
A
ut
hor
s
s
ta
te
no f
undi
ng i
nvol
ve
d.
A
U
T
H
O
R
C
O
N
T
R
I
B
U
T
I
O
N
S
S
T
A
T
E
M
E
N
T
T
hi
s
jo
ur
na
l
us
e
s
th
e
C
ont
r
ib
ut
or
R
ol
e
s
T
a
xonomy
(
C
R
e
di
T
)
to
r
e
c
ogni
z
e
in
di
vi
dua
l
a
ut
hor
c
ont
r
ib
ut
io
ns
, r
e
duc
e
a
ut
hor
s
hi
p di
s
put
e
s
,
a
nd f
a
c
il
it
a
te
c
ol
la
bo
r
a
ti
on.
N
am
e
o
f
A
u
t
h
or
C
M
So
Va
Fo
I
R
D
O
E
Vi
Su
P
Fu
T
h
a
k
w
a
n
A
kr
a
m
J
a
w
a
d
✓
✓
✓
✓
✓
✓
✓
✓
J
a
m
s
hi
d
B
a
ghe
r
z
a
d
e
h
M
oha
s
e
f
i
✓
✓
✓
✓
✓
✓
M
o
ha
m
m
e
d S
a
la
h R
e
d
a
A
bd
e
l
gh
a
ny
✓
✓
✓
✓
✓
C
:
C
onc
e
pt
ua
l
i
z
a
t
i
on
M
:
M
e
t
hodol
ogy
So
:
So
f
t
w
a
r
e
Va
:
Va
l
i
da
t
i
on
Fo
:
Fo
r
m
a
l
a
na
l
ys
i
s
I
:
I
nve
s
t
i
ga
t
i
on
R
:
R
e
s
our
c
e
s
D
:
D
a
t
a
C
ur
a
t
i
on
O
:
W
r
i
t
i
ng
-
O
r
i
gi
na
l
D
r
a
f
t
E
:
W
r
i
t
i
ng
-
R
e
vi
e
w
&
E
di
t
i
ng
Vi
:
Vi
s
ua
l
i
z
a
t
i
on
Su
:
Su
pe
r
vi
s
i
on
P
:
P
r
oj
e
c
t
a
dm
i
ni
s
t
r
a
t
i
on
Fu
:
Fu
ndi
ng a
c
qui
s
i
t
i
on
Evaluation Warning : The document was created with Spire.PDF for Python.